mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-04-19 22:39:03 +00:00
Grouped convolution backward data WMMA v3 implementation (#3460)
* Added device level implementation for bwd_data_wmma_v3. * Added first instance of bwd_data_wmma_v3(f16). * Add support for bwd data in gridwise implementation Some changes are general for convolution and some are specific for bwd data. We need to generalize them once we have fwd, bwd data and bwd weight * Initial device implementation of bwd data * Remove unused template parameters in device impl * Add one instance for different layout initial check of device implementation * Add tests for splitk and for different layouts * Appended more instances to wmma_v3_f16. * Added conv_2d bf16 wmma_v3 instances. * Added conv_3d_bf16 wmma_v3_instances. * Added conv_3d_f16_wmma_v3_instances. * Added SplitN test cases for wmma. * Conv3d_bwd_data_scale_wmma_v3 instances. * Conv3d_bwd_data_bilinear_wmma_v3_instances * Renaming the device level instances file to common name , since it is defined for different DataTypes. * Renaming the instances and fixing typo * Added the test cases to regression test list * NCHW support for wmma_v3 * Examples for bf16 and f16 bwd_data_wmma_v3 * Added transpose conditons for device impl * fixing bugs * Added the gemm_args array implmentation * WIP debug conv bwd * fix splitk * Grouped gemm fix * Update CmakeLists with EOF * Added more instances for tests * Fixed the run time error in examples and removed 3d conv examples. * Fixed a typo. * Updated CmakeLists to removed the 3d convultion deleted files * Added print error statements for unsupoorted argument * Added the merge conflict related changes * Fixed compilation error * Fixed the InstanceFactory duplication error. * Removed the print statements and added logs to Arg function * All the merge conflict related errors resolved * Added d_tensor tests. * Added the missing example types of wmm_v3 * Merge error fix * Corrected the instance name * Reverted the bias relu change * Revereted the transpose load local change * Updated the regression test list with bwd_data_scale * Revert "Revereted the transpose load local change" This reverts commit 0b7281edb2bf008e407006690a00621174d9d19b. * Revert "Merge error fix" This reverts commit f3c85daa474b1b83d10c8a3ce077354e71d91a2b. * Reverting the local change * Added merge error fix * Build error fix due to merge conflicts * Added bias_relu example for wmma_v3 * Modified the main method in dtensor tests * Updated the dtensor tests to pick all the shapes * Updated the dtensor test shapes. * Updated the mem operations in tests. * Added reference func * Fixed typos in device impl * Added new header file and modified the include file for 3d tests * Renamed the test file and added reference func call. * clang format fix * Added ignore params * Modified device impl and tests * Removed debug print statements and updated dtensor test shapes * Fixing merge conflicts * Fixing more merge conflicts * Fixed copyrights * Updated the tuned instances to bilinear and scale. * Adding tuned instances to vanilla wmma_v3 * Removed all unused instances and modified test layouts. * Cleaned up all instances , reverted back fwd fp16 instances and updated tuned fp16 instances. * Fix clang format * Updated tuned f16/-genric instances * Formatting the instances file * Fixed copyrights and clang issues * Nonsense commit to force git to force * Removed the transpose instances * Added verified genric instances * Fixing namespace errors * Added todo for failing shapes * Formatting instance file * Fix instance list formatting * Removing unnecessary formats * Renamed the common file * Unification of xdl and wmma bwd_data tests * Updated Cmake * Added all layout types and deleted code. * Updated Cmake to add the condition to all tests. --------- Co-authored-by: Enrico Degregori <enrico@streamhpc.com> Co-authored-by: Anton Gorenko <anton@streamhpc.com> Co-authored-by: kiefer <kiefer.van.teutem@streamhpc.com>
This commit is contained in:
ApoorvaKalyani
@@ -40,6 +40,9 @@ set(REGRESSION_TESTS
|
||||
test_batchnorm_fwd_rank_4
|
||||
test_batchnorm_bwd_rank_4
|
||||
test_grouped_convnd_bwd_data_xdl
|
||||
test_grouped_convnd_bwd_data_wmma
|
||||
test_grouped_convnd_bwd_data_wmma_large_cases
|
||||
test_grouped_conv_bwd_data_wmma_scale
|
||||
test_conv_tensor_rearrange
|
||||
test_gemm_mx
|
||||
test_ck_tile_batched_transpose
|
||||
|
||||
@@ -1,22 +1,25 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_gtest_executable(test_grouped_convnd_bwd_data_xdl test_grouped_convnd_bwd_data_xdl.cpp)
|
||||
if(result EQUAL 0)
|
||||
target_link_libraries(test_grouped_convnd_bwd_data_xdl PRIVATE utility device_grouped_conv2d_bwd_data_instance device_grouped_conv3d_bwd_data_instance)
|
||||
endif()
|
||||
if(GPU_TARGETS MATCHES "gfx9|gfx11|gfx12")
|
||||
add_executable(test_grouped_convnd_bwd_data_xdl_large_cases test_grouped_convnd_bwd_data_xdl_large_cases.cpp)
|
||||
target_compile_options(test_grouped_convnd_bwd_data_xdl_large_cases PRIVATE -Wno-global-constructors -Wno-undef)
|
||||
target_link_libraries(test_grouped_convnd_bwd_data_xdl_large_cases PRIVATE gtest_main getopt::getopt utility device_grouped_conv2d_bwd_data_instance device_grouped_conv3d_bwd_data_instance)
|
||||
add_gtest_executable(test_grouped_convnd_bwd_data test_grouped_convnd_bwd_data.cpp)
|
||||
target_link_libraries(test_grouped_convnd_bwd_data PRIVATE utility device_grouped_conv2d_bwd_data_instance device_grouped_conv3d_bwd_data_instance)
|
||||
|
||||
add_executable(test_grouped_convnd_bwd_data_large_cases test_grouped_convnd_bwd_data_large_cases.cpp)
|
||||
target_compile_options(test_grouped_convnd_bwd_data_large_cases PRIVATE -Wno-global-constructors -Wno-undef)
|
||||
target_link_libraries(test_grouped_convnd_bwd_data_large_cases PRIVATE gtest_main getopt::getopt utility device_grouped_conv2d_bwd_data_instance device_grouped_conv3d_bwd_data_instance)
|
||||
|
||||
add_executable(test_grouped_convnd_bwd_data_dataset_xdl test_grouped_convnd_bwd_data_dataset_xdl.cpp)
|
||||
target_compile_options(test_grouped_convnd_bwd_data_dataset_xdl PRIVATE -Wno-global-constructors -Wno-undef)
|
||||
target_link_libraries(test_grouped_convnd_bwd_data_dataset_xdl PRIVATE gtest_main getopt::getopt utility device_grouped_conv2d_bwd_data_instance device_grouped_conv3d_bwd_data_instance)
|
||||
endif()
|
||||
add_gtest_executable(test_grouped_convnd_bwd_data_wmma test_grouped_convnd_bwd_data_wmma.cpp)
|
||||
if(result EQUAL 0)
|
||||
target_link_libraries(test_grouped_convnd_bwd_data_wmma PRIVATE utility device_grouped_conv2d_bwd_data_instance device_grouped_conv3d_bwd_data_instance)
|
||||
|
||||
add_executable(test_grouped_conv_bwd_data_bilinear test_grouped_conv_bwd_data_bilinear.cpp)
|
||||
target_compile_options(test_grouped_conv_bwd_data_bilinear PRIVATE -Wno-global-constructors -Wno-undef)
|
||||
target_link_libraries(test_grouped_conv_bwd_data_bilinear PRIVATE gtest_main getopt::getopt utility device_grouped_conv3d_bwd_data_bilinear_instance)
|
||||
|
||||
add_executable(test_grouped_conv_bwd_data_scale test_grouped_conv_bwd_data_scale.cpp)
|
||||
target_compile_options(test_grouped_conv_bwd_data_scale PRIVATE -Wno-global-constructors -Wno-undef)
|
||||
target_link_libraries(test_grouped_conv_bwd_data_scale PRIVATE gtest_main getopt::getopt utility device_grouped_conv3d_bwd_data_scale_instance)
|
||||
endif()
|
||||
add_gtest_executable(test_grouped_convnd_bwd_data_interface_xdl test_grouped_convnd_bwd_data_interface_xdl.cpp)
|
||||
if(result EQUAL 0)
|
||||
@@ -25,4 +28,4 @@ endif()
|
||||
add_gtest_executable(test_grouped_convnd_bwd_data_interface_wmma test_grouped_convnd_bwd_data_interface_wmma.cpp)
|
||||
if(result EQUAL 0)
|
||||
target_link_libraries(test_grouped_convnd_bwd_data_interface_wmma PRIVATE utility device_grouped_conv2d_bwd_data_instance)
|
||||
endif()
|
||||
endif()
|
||||
@@ -0,0 +1,324 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#include <algorithm>
|
||||
#include <iomanip>
|
||||
#include <iostream>
|
||||
#include <iterator>
|
||||
#include <typeinfo>
|
||||
#include <gtest/gtest.h>
|
||||
|
||||
#include "ck/ck.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
|
||||
|
||||
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
|
||||
|
||||
#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data_bilinear.hpp"
|
||||
|
||||
#include "ck/library/utility/check_err.hpp"
|
||||
#include "ck/library/utility/device_memory.hpp"
|
||||
#include "ck/library/utility/host_tensor.hpp"
|
||||
#include "ck/library/utility/host_tensor_generator.hpp"
|
||||
#include "ck/library/utility/convolution_parameter.hpp"
|
||||
#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
|
||||
#include "ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp"
|
||||
|
||||
using ::ck::DeviceMem;
|
||||
using ::ck::HostTensorDescriptor;
|
||||
using ::ck::Tensor;
|
||||
|
||||
static ck::index_t param_mask = 0xffff;
|
||||
static ck::index_t instance_index = -1;
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdData : public ::testing::Test
|
||||
{
|
||||
protected:
|
||||
using InDataType = std::tuple_element_t<0, Tuple>;
|
||||
using WeiDataType = std::tuple_element_t<0, Tuple>;
|
||||
using OutDataType = std::tuple_element_t<0, Tuple>;
|
||||
|
||||
using ComputeDataType = InDataType;
|
||||
using InLayout = std::tuple_element_t<3, Tuple>;
|
||||
using WeiLayout = std::tuple_element_t<2, Tuple>;
|
||||
using OutLayout = std::tuple_element_t<1, Tuple>;
|
||||
|
||||
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using InElementOp = ck::tensor_operation::element_wise::Bilinear;
|
||||
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
using Bilinear = ck::tensor_operation::element_wise::Bilinear;
|
||||
static constexpr ck::index_t NDimSpatial = 3;
|
||||
static constexpr float alpha = 2.f;
|
||||
static constexpr float beta = 2.f;
|
||||
static constexpr ck::index_t NumDs = 1;
|
||||
|
||||
std::vector<ck::utils::conv::ConvParam> conv_params;
|
||||
std::vector<ck::index_t> split_ks{1};
|
||||
|
||||
void RunReference(ck::utils::conv::ConvParam& conv_param,
|
||||
Tensor<InDataType>& in_host,
|
||||
Tensor<WeiDataType>& wei,
|
||||
Tensor<OutDataType>& out,
|
||||
Tensor<InDataType>& d)
|
||||
{
|
||||
|
||||
std::array<Tensor<InDataType>, NumDs> d_tensors = {d};
|
||||
auto ref_conv =
|
||||
ck::tensor_operation::host::ReferenceConvBwdData<NDimSpatial,
|
||||
InDataType,
|
||||
WeiDataType,
|
||||
OutDataType,
|
||||
InElementOp,
|
||||
WeiElementOp,
|
||||
OutElementOp,
|
||||
0, /*Num A Elementwise Tensors*/
|
||||
0, /*Num B Elementwise Tensors*/
|
||||
NumDs>();
|
||||
|
||||
auto ref_invoker = ref_conv.MakeInvoker();
|
||||
|
||||
auto ref_argument = ref_conv.MakeArgument(in_host,
|
||||
wei,
|
||||
out,
|
||||
conv_param.conv_filter_strides_,
|
||||
conv_param.conv_filter_dilations_,
|
||||
conv_param.input_left_pads_,
|
||||
conv_param.input_right_pads_,
|
||||
Bilinear{alpha, beta},
|
||||
WeiElementOp{},
|
||||
OutElementOp{},
|
||||
{},
|
||||
{},
|
||||
d_tensors);
|
||||
|
||||
ref_invoker.Run(ref_argument);
|
||||
}
|
||||
|
||||
bool PerformConvDataBilinear(ck::utils::conv::ConvParam& conv_param,
|
||||
const ck::index_t split_k,
|
||||
ck::index_t instance_index_ = -1)
|
||||
{
|
||||
bool passed = true;
|
||||
|
||||
const auto out_g_n_k_wos_desc =
|
||||
ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
|
||||
conv_param);
|
||||
|
||||
const auto wei_g_k_c_xs_desc =
|
||||
ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
|
||||
conv_param);
|
||||
|
||||
const auto in_g_n_c_wis_desc =
|
||||
ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
|
||||
conv_param);
|
||||
|
||||
Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
|
||||
Tensor<OutDataType> out(out_g_n_k_wos_desc);
|
||||
Tensor<InDataType> in_host(in_g_n_c_wis_desc);
|
||||
Tensor<InDataType> in_device(in_g_n_c_wis_desc);
|
||||
Tensor<InDataType> d(in_g_n_c_wis_desc);
|
||||
|
||||
std::cout << "in: " << in_host.mDesc << std::endl;
|
||||
std::cout << "wei: " << wei.mDesc << std::endl;
|
||||
std::cout << "out: " << out.mDesc << std::endl;
|
||||
|
||||
out.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
|
||||
wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
|
||||
d.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5});
|
||||
|
||||
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
|
||||
DeviceMem out_device_buf(sizeof(OutDataType) * out.mDesc.GetElementSpaceSize());
|
||||
DeviceMem in_device_buf(sizeof(InDataType) * in_device.mDesc.GetElementSpaceSize());
|
||||
DeviceMem d_device_buf(sizeof(InDataType) * d.mDesc.GetElementSpaceSize());
|
||||
|
||||
in_device_buf.ToDevice(in_device.mData.data());
|
||||
out_device_buf.ToDevice(out.mData.data());
|
||||
wei_device_buf.ToDevice(wei.mData.data());
|
||||
d_device_buf.ToDevice(d.mData.data());
|
||||
|
||||
std::array<ck::index_t, NDimSpatial + 3> out_lengths{};
|
||||
std::array<ck::index_t, NDimSpatial + 3> out_strides{};
|
||||
std::array<ck::index_t, NDimSpatial + 3> wei_lengths{};
|
||||
std::array<ck::index_t, NDimSpatial + 3> wei_strides{};
|
||||
std::array<ck::index_t, NDimSpatial + 3> in_lengths{};
|
||||
std::array<ck::index_t, NDimSpatial + 3> in_strides{};
|
||||
std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
|
||||
std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
|
||||
std::array<ck::index_t, NDimSpatial> input_left_pads{};
|
||||
std::array<ck::index_t, NDimSpatial> input_right_pads{};
|
||||
|
||||
auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
|
||||
|
||||
copy(out_g_n_k_wos_desc.GetLengths(), out_lengths);
|
||||
copy(out_g_n_k_wos_desc.GetStrides(), out_strides);
|
||||
copy(wei_g_k_c_xs_desc.GetLengths(), wei_lengths);
|
||||
copy(wei_g_k_c_xs_desc.GetStrides(), wei_strides);
|
||||
copy(in_g_n_c_wis_desc.GetLengths(), in_lengths);
|
||||
copy(in_g_n_c_wis_desc.GetStrides(), in_strides);
|
||||
copy(conv_param.conv_filter_strides_, conv_filter_strides);
|
||||
copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
|
||||
copy(conv_param.input_left_pads_, input_left_pads);
|
||||
copy(conv_param.input_right_pads_, input_right_pads);
|
||||
|
||||
RunReference(conv_param, in_host, wei, out, d);
|
||||
|
||||
using DeviceOp =
|
||||
ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<NDimSpatial,
|
||||
OutLayout,
|
||||
WeiLayout,
|
||||
ck::Tuple<InLayout>,
|
||||
InLayout,
|
||||
OutDataType,
|
||||
WeiDataType,
|
||||
ck::Tuple<InDataType>,
|
||||
InDataType,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
Bilinear>;
|
||||
|
||||
// get device op instances
|
||||
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
|
||||
DeviceOp>::GetInstances();
|
||||
|
||||
int num_kernel = 0;
|
||||
|
||||
for(std::size_t i = 0; i < op_ptrs.size(); ++i)
|
||||
{
|
||||
auto& op_ptr = op_ptrs[i];
|
||||
auto argument_ptr = op_ptr->MakeArgumentPointer(
|
||||
out_device_buf.GetDeviceBuffer(),
|
||||
wei_device_buf.GetDeviceBuffer(),
|
||||
{d_device_buf.GetDeviceBuffer()},
|
||||
in_device_buf.GetDeviceBuffer(),
|
||||
out_lengths,
|
||||
out_strides,
|
||||
wei_lengths,
|
||||
wei_strides,
|
||||
std::array<std::array<ck::index_t, NDimSpatial + 3>, NumDs>{in_lengths},
|
||||
std::array<std::array<ck::index_t, NDimSpatial + 3>, NumDs>{in_strides},
|
||||
in_lengths,
|
||||
in_strides,
|
||||
conv_filter_strides,
|
||||
conv_filter_dilations,
|
||||
input_left_pads,
|
||||
input_right_pads,
|
||||
PassThrough{},
|
||||
PassThrough{},
|
||||
Bilinear{alpha, beta},
|
||||
split_k);
|
||||
|
||||
DeviceMem workspace_buf(op_ptr->GetWorkSpaceSize(argument_ptr.get()));
|
||||
op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_buf.GetDeviceBuffer());
|
||||
|
||||
auto invoker_ptr = op_ptr->MakeInvokerPointer();
|
||||
std::string op_name = op_ptr->GetTypeString();
|
||||
|
||||
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
|
||||
{
|
||||
++num_kernel;
|
||||
if((instance_index_ != -1) && (instance_index_ + 1 != num_kernel))
|
||||
{
|
||||
// skip test if instance_index is specified
|
||||
continue;
|
||||
}
|
||||
|
||||
float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
|
||||
in_device_buf.FromDevice(in_device.mData.data());
|
||||
|
||||
passed &= ck::utils::check_err(in_device, in_host);
|
||||
|
||||
std::size_t flop = conv_param.GetFlops() +
|
||||
3 * conv_param.GetOutputByte<InDataType>() / sizeof(InDataType);
|
||||
std::size_t num_bytes = conv_param.GetByte<InDataType, WeiDataType, OutDataType>() +
|
||||
conv_param.GetOutputByte<InDataType>();
|
||||
|
||||
float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
|
||||
float gb_per_sec = num_bytes / 1.E6 / avg_time;
|
||||
|
||||
std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops
|
||||
<< " TFlops, " << gb_per_sec << " GB/s, " << op_name << std::endl;
|
||||
}
|
||||
else
|
||||
{
|
||||
std::cerr << op_name << " does not support this problem" << std::endl;
|
||||
}
|
||||
}
|
||||
if(instance_index != -1)
|
||||
{
|
||||
std::cout << "grouped_conv_bwd_data_instance (" << instance_index << "/" << num_kernel
|
||||
<< "): Passed" << std::endl;
|
||||
}
|
||||
printf("\033[36mvalids: %d\033[0m\n", num_kernel);
|
||||
return passed;
|
||||
}
|
||||
|
||||
void Run()
|
||||
{
|
||||
EXPECT_FALSE(conv_params.empty());
|
||||
bool pass = true;
|
||||
for(auto split_k : split_ks)
|
||||
{
|
||||
for(auto& param : conv_params)
|
||||
{
|
||||
pass = pass && PerformConvDataBilinear(param, split_k, instance_index);
|
||||
}
|
||||
}
|
||||
EXPECT_TRUE(pass);
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdData3d : public TestGroupedConvndBwdData<Tuple>
|
||||
{
|
||||
};
|
||||
|
||||
using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
|
||||
using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
|
||||
using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
|
||||
|
||||
using KernelTypes3d = ::testing::Types<std::tuple<float, NDHWGK, GKZYXC, NDHWGC>,
|
||||
std::tuple<ck::half_t, NDHWGK, GKZYXC, NDHWGC>,
|
||||
std::tuple<ck::bhalf_t, NDHWGK, GKZYXC, NDHWGC>>;
|
||||
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdData3d, KernelTypes3d);
|
||||
|
||||
TYPED_TEST(TestGroupedConvndBwdData3d, Test3D)
|
||||
{
|
||||
// TODO: To fix the impl to pass with stride greater than 1.
|
||||
// this->conv_params.push_back(
|
||||
// {3, 2, 16, 128, 128, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
|
||||
this->conv_params.push_back(
|
||||
{3, 2, 16, 128, 128, {1, 1, 1}, {7, 7, 7}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
|
||||
this->conv_params.push_back(
|
||||
{3, 2, 2, 128, 128, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 2, 32, 128, 128, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 1, 32, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 64, 3, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 1, 1, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 4, 4, {3, 3, 3}, {14, 28, 28}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->Run();
|
||||
}
|
||||
|
||||
int main(int argc, char** argv)
|
||||
{
|
||||
testing::InitGoogleTest(&argc, argv);
|
||||
if(argc == 1) {}
|
||||
else if(argc == 3)
|
||||
{
|
||||
param_mask = strtol(argv[1], nullptr, 0);
|
||||
instance_index = atoi(argv[2]);
|
||||
}
|
||||
else
|
||||
{
|
||||
std::cout << "Usage of " << argv[0] << std::endl;
|
||||
std::cout << "Arg1,2: param_mask instance_index(-1 means all)" << std::endl;
|
||||
}
|
||||
return RUN_ALL_TESTS();
|
||||
}
|
||||
@@ -0,0 +1,324 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#include <algorithm>
|
||||
#include <iomanip>
|
||||
#include <iostream>
|
||||
#include <iterator>
|
||||
#include <typeinfo>
|
||||
#include <gtest/gtest.h>
|
||||
|
||||
#include "ck/ck.hpp"
|
||||
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
|
||||
|
||||
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
|
||||
|
||||
#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data_scale.hpp"
|
||||
|
||||
#include "ck/library/utility/check_err.hpp"
|
||||
#include "ck/library/utility/device_memory.hpp"
|
||||
#include "ck/library/utility/host_tensor.hpp"
|
||||
#include "ck/library/utility/host_tensor_generator.hpp"
|
||||
#include "ck/library/utility/convolution_parameter.hpp"
|
||||
#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
|
||||
#include "ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp"
|
||||
|
||||
using ::ck::DeviceMem;
|
||||
using ::ck::HostTensorDescriptor;
|
||||
using ::ck::Tensor;
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdData : public ::testing::Test
|
||||
{
|
||||
protected:
|
||||
using F16 = ck::half_t;
|
||||
using InDataType = std::tuple_element_t<0, Tuple>;
|
||||
using WeiDataType = std::tuple_element_t<0, Tuple>;
|
||||
using OutDataType = std::tuple_element_t<0, Tuple>;
|
||||
|
||||
using ComputeDataType = InDataType;
|
||||
using InLayout = std::tuple_element_t<3, Tuple>;
|
||||
using WeiLayout = std::tuple_element_t<2, Tuple>;
|
||||
using OutLayout = std::tuple_element_t<1, Tuple>;
|
||||
|
||||
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using InElementOp = ck::tensor_operation::element_wise::Scale;
|
||||
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
using Scale = ck::tensor_operation::element_wise::Scale;
|
||||
static constexpr ck::index_t NDimSpatial = 3;
|
||||
static constexpr float alpha = 2.f;
|
||||
|
||||
std::vector<ck::utils::conv::ConvParam> conv_params;
|
||||
std::vector<ck::index_t> split_ks{1};
|
||||
|
||||
void RunReference(ck::utils::conv::ConvParam& conv_param,
|
||||
Tensor<InDataType>& in_host,
|
||||
Tensor<WeiDataType>& wei,
|
||||
Tensor<OutDataType>& out)
|
||||
{
|
||||
auto ref_conv =
|
||||
ck::tensor_operation::host::ReferenceConvBwdData<NDimSpatial,
|
||||
InDataType,
|
||||
WeiDataType,
|
||||
OutDataType,
|
||||
InElementOp,
|
||||
WeiElementOp,
|
||||
OutElementOp,
|
||||
0, /*Num A Elementwise Tensors*/
|
||||
0, /*Num B Elementwise Tensors*/
|
||||
0,
|
||||
ComputeDataType> /*Num D Elementwise
|
||||
Tensors*/
|
||||
{};
|
||||
|
||||
auto ref_invoker = ref_conv.MakeInvoker();
|
||||
|
||||
auto ref_argument = ref_conv.MakeArgument(in_host,
|
||||
wei,
|
||||
out,
|
||||
conv_param.conv_filter_strides_,
|
||||
conv_param.conv_filter_dilations_,
|
||||
conv_param.input_left_pads_,
|
||||
conv_param.input_right_pads_,
|
||||
InElementOp{alpha},
|
||||
WeiElementOp{},
|
||||
OutElementOp{});
|
||||
|
||||
ref_invoker.Run(ref_argument);
|
||||
}
|
||||
|
||||
bool PerformConvDataScale(ck::utils::conv::ConvParam& conv_param, const ck::index_t split_k)
|
||||
{
|
||||
bool passed = true;
|
||||
|
||||
const auto out_g_n_k_wos_desc =
|
||||
ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
|
||||
conv_param);
|
||||
|
||||
const auto wei_g_k_c_xs_desc =
|
||||
ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
|
||||
conv_param);
|
||||
|
||||
const auto in_g_n_c_wis_desc =
|
||||
ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
|
||||
conv_param);
|
||||
|
||||
Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
|
||||
Tensor<OutDataType> out(out_g_n_k_wos_desc);
|
||||
Tensor<InDataType> in_host(in_g_n_c_wis_desc);
|
||||
Tensor<InDataType> in_device(in_g_n_c_wis_desc);
|
||||
|
||||
std::cout << "in: " << in_host.mDesc << std::endl;
|
||||
std::cout << "wei: " << wei.mDesc << std::endl;
|
||||
std::cout << "out: " << out.mDesc << std::endl;
|
||||
|
||||
out.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
|
||||
wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
|
||||
|
||||
DeviceMem in_device_buf(sizeof(InDataType) * in_device.mDesc.GetElementSpaceSize());
|
||||
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
|
||||
DeviceMem out_device_buf(sizeof(OutDataType) * out.mDesc.GetElementSpaceSize());
|
||||
|
||||
in_device_buf.ToDevice(in_device.mData.data());
|
||||
out_device_buf.ToDevice(out.mData.data());
|
||||
wei_device_buf.ToDevice(wei.mData.data());
|
||||
|
||||
std::array<ck::index_t, NDimSpatial + 3> out_lengths{};
|
||||
std::array<ck::index_t, NDimSpatial + 3> out_strides{};
|
||||
std::array<ck::index_t, NDimSpatial + 3> wei_lengths{};
|
||||
std::array<ck::index_t, NDimSpatial + 3> wei_strides{};
|
||||
std::array<ck::index_t, NDimSpatial + 3> in_lengths{};
|
||||
std::array<ck::index_t, NDimSpatial + 3> in_strides{};
|
||||
std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
|
||||
std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
|
||||
std::array<ck::index_t, NDimSpatial> input_left_pads{};
|
||||
std::array<ck::index_t, NDimSpatial> input_right_pads{};
|
||||
|
||||
auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
|
||||
|
||||
copy(out_g_n_k_wos_desc.GetLengths(), out_lengths);
|
||||
copy(out_g_n_k_wos_desc.GetStrides(), out_strides);
|
||||
copy(wei_g_k_c_xs_desc.GetLengths(), wei_lengths);
|
||||
copy(wei_g_k_c_xs_desc.GetStrides(), wei_strides);
|
||||
copy(in_g_n_c_wis_desc.GetLengths(), in_lengths);
|
||||
copy(in_g_n_c_wis_desc.GetStrides(), in_strides);
|
||||
copy(conv_param.conv_filter_strides_, conv_filter_strides);
|
||||
copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
|
||||
copy(conv_param.input_left_pads_, input_left_pads);
|
||||
copy(conv_param.input_right_pads_, input_right_pads);
|
||||
|
||||
RunReference(conv_param, in_host, wei, out);
|
||||
|
||||
using DeviceOp =
|
||||
ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<NDimSpatial,
|
||||
OutLayout,
|
||||
WeiLayout,
|
||||
ck::Tuple<>,
|
||||
InLayout,
|
||||
OutDataType,
|
||||
WeiDataType,
|
||||
ck::Tuple<>,
|
||||
InDataType,
|
||||
PassThrough,
|
||||
PassThrough,
|
||||
Scale>;
|
||||
|
||||
// get device op instances
|
||||
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
|
||||
DeviceOp>::GetInstances();
|
||||
|
||||
int num_kernel = 0;
|
||||
|
||||
for(std::size_t i = 0; i < op_ptrs.size(); ++i)
|
||||
{
|
||||
auto& op_ptr = op_ptrs[i];
|
||||
auto argument_ptr = op_ptr->MakeArgumentPointer(out_device_buf.GetDeviceBuffer(),
|
||||
wei_device_buf.GetDeviceBuffer(),
|
||||
{},
|
||||
in_device_buf.GetDeviceBuffer(),
|
||||
out_lengths,
|
||||
out_strides,
|
||||
wei_lengths,
|
||||
wei_strides,
|
||||
{},
|
||||
{},
|
||||
in_lengths,
|
||||
in_strides,
|
||||
conv_filter_strides,
|
||||
conv_filter_dilations,
|
||||
input_left_pads,
|
||||
input_right_pads,
|
||||
PassThrough{},
|
||||
PassThrough{},
|
||||
Scale{alpha});
|
||||
|
||||
DeviceMem workspace_buf(op_ptr->GetWorkSpaceSize(argument_ptr.get()));
|
||||
op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_buf.GetDeviceBuffer());
|
||||
|
||||
auto invoker_ptr = op_ptr->MakeInvokerPointer();
|
||||
std::string op_name = op_ptr->GetTypeString();
|
||||
|
||||
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
|
||||
{
|
||||
num_kernel++;
|
||||
float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
|
||||
in_device_buf.FromDevice(in_device.mData.data());
|
||||
|
||||
using ComputeType_ = std::conditional_t<sizeof(OutDataType) < sizeof(InDataType),
|
||||
OutDataType,
|
||||
InDataType>;
|
||||
using ComputeType =
|
||||
std::conditional_t<sizeof(ComputeType_) < sizeof(ComputeDataType),
|
||||
ComputeType_,
|
||||
ComputeDataType>;
|
||||
using AccDataType =
|
||||
std::conditional_t<std::is_same_v<ComputeType, int8_t>, int32_t, float>;
|
||||
const ck::index_t num_accums = conv_param.K_;
|
||||
float max_accumulated_value =
|
||||
*std::max_element(in_host.mData.begin(), in_host.mData.end());
|
||||
|
||||
const ck::index_t split_k_for_run = split_k;
|
||||
// Calculate thresholds
|
||||
auto rtol = ck::utils::get_relative_threshold<ComputeType, InDataType, AccDataType>(
|
||||
num_accums / split_k_for_run);
|
||||
auto atol = ck::utils::get_absolute_threshold<ComputeType, InDataType, AccDataType>(
|
||||
max_accumulated_value / split_k_for_run, num_accums / split_k_for_run);
|
||||
// Calculate error due to split_k accumulation
|
||||
auto rtol_split_k =
|
||||
ck::utils::get_relative_threshold<InDataType, InDataType, InDataType>(
|
||||
split_k_for_run);
|
||||
auto atol_split_k =
|
||||
ck::utils::get_absolute_threshold<InDataType, InDataType, InDataType>(
|
||||
max_accumulated_value, split_k_for_run);
|
||||
// Use higher threshold
|
||||
rtol = std::max(rtol, rtol_split_k);
|
||||
atol = std::max(atol, atol_split_k);
|
||||
if(split_k_for_run > 1)
|
||||
{
|
||||
passed &= ck::utils::check_err(
|
||||
in_device, in_host, "Error: Incorrect results!", rtol, atol);
|
||||
std::cout << "Relative error threshold: " << rtol
|
||||
<< " Absolute error threshold: " << atol << std::endl;
|
||||
}
|
||||
else
|
||||
{
|
||||
passed &= ck::utils::check_err(
|
||||
in_device, in_host, "Error: Incorrect results!", rtol, atol);
|
||||
std::cout << "Relative error threshold: " << rtol
|
||||
<< " Absolute error threshold: " << atol << std::endl;
|
||||
}
|
||||
std::size_t flop = conv_param.GetFlops() +
|
||||
3 * conv_param.GetOutputByte<InDataType>() / sizeof(InDataType);
|
||||
std::size_t num_bytes = conv_param.GetByte<InDataType, WeiDataType, OutDataType>() +
|
||||
conv_param.GetOutputByte<InDataType>();
|
||||
|
||||
float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
|
||||
float gb_per_sec = num_bytes / 1.E6 / avg_time;
|
||||
|
||||
std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops
|
||||
<< " TFlops, " << gb_per_sec << " GB/s, " << op_name << std::endl;
|
||||
}
|
||||
else
|
||||
{
|
||||
std::cerr << op_name << " does not support this problem" << std::endl;
|
||||
}
|
||||
}
|
||||
|
||||
printf("\033[36mvalids: %d\033[0m\n", num_kernel);
|
||||
return passed;
|
||||
}
|
||||
|
||||
void Run()
|
||||
{
|
||||
EXPECT_FALSE(conv_params.empty());
|
||||
bool pass = true;
|
||||
|
||||
for(auto split_k : split_ks)
|
||||
{
|
||||
for(auto& param : conv_params)
|
||||
{
|
||||
pass = pass && PerformConvDataScale(param, split_k);
|
||||
}
|
||||
}
|
||||
EXPECT_TRUE(pass);
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdData3d : public TestGroupedConvndBwdData<Tuple>
|
||||
{
|
||||
};
|
||||
|
||||
using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
|
||||
using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
|
||||
using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
|
||||
|
||||
using KernelTypes3d = ::testing::Types<std::tuple<float, NDHWGK, GKZYXC, NDHWGC>,
|
||||
std::tuple<ck::half_t, NDHWGK, GKZYXC, NDHWGC>,
|
||||
std::tuple<ck::bhalf_t, NDHWGK, GKZYXC, NDHWGC>>;
|
||||
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdData3d, KernelTypes3d);
|
||||
|
||||
TYPED_TEST(TestGroupedConvndBwdData3d, Test3D)
|
||||
{
|
||||
this->conv_params.push_back(
|
||||
{3, 2, 16, 128, 128, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
|
||||
this->conv_params.push_back(
|
||||
{3, 2, 2, 128, 128, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 2, 32, 128, 128, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 1, 32, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 64, 3, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 1, 1, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 4, 4, {3, 3, 3}, {14, 28, 28}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 64, 16, 32, {3, 3, 3}, {28, 28, 28}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
|
||||
this->Run();
|
||||
}
|
||||
@@ -15,7 +15,7 @@ static ck::index_t param_mask = 0xffffff;
|
||||
static ck::index_t instance_index = -1;
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdDataXdl : public ::testing::Test
|
||||
class TestGroupedConvndBwdData : public ::testing::Test
|
||||
{
|
||||
protected:
|
||||
using DataType = std::tuple_element_t<0, Tuple>;
|
||||
@@ -89,19 +89,19 @@ using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWK, GKZYXC, GNDHWC>
|
||||
std::tuple<ck::bhalf_t, NDHWGK, GKZYXC, NDHWGC>>;
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdDataXdl2d : public TestGroupedConvndBwdDataXdl<Tuple>
|
||||
class TestGroupedConvndBwdData2d : public TestGroupedConvndBwdData<Tuple>
|
||||
{
|
||||
};
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdDataXdl3d : public TestGroupedConvndBwdDataXdl<Tuple>
|
||||
class TestGroupedConvndBwdData3d : public TestGroupedConvndBwdData<Tuple>
|
||||
{
|
||||
};
|
||||
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdDataXdl2d, KernelTypes2d);
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdDataXdl3d, KernelTypes3d);
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdData2d, KernelTypes2d);
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdData3d, KernelTypes3d);
|
||||
|
||||
TYPED_TEST(TestGroupedConvndBwdDataXdl2d, Test2D)
|
||||
TYPED_TEST(TestGroupedConvndBwdData2d, Test2D)
|
||||
{
|
||||
this->conv_params.clear();
|
||||
|
||||
@@ -137,7 +137,7 @@ TYPED_TEST(TestGroupedConvndBwdDataXdl2d, Test2D)
|
||||
this->template Run<2>();
|
||||
}
|
||||
|
||||
TYPED_TEST(TestGroupedConvndBwdDataXdl3d, Test3D)
|
||||
TYPED_TEST(TestGroupedConvndBwdData3d, Test3D)
|
||||
{
|
||||
this->conv_params.clear();
|
||||
this->conv_params.push_back(
|
||||
@@ -12,7 +12,7 @@
|
||||
#include "profiler/profile_grouped_conv_bwd_data_impl.hpp"
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdDataXdl : public ::testing::Test
|
||||
class TestGroupedConvndBwdData : public ::testing::Test
|
||||
{
|
||||
protected:
|
||||
using DataType = std::tuple_element_t<0, Tuple>;
|
||||
@@ -80,19 +80,19 @@ using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWK, GKZYXC, GNDHWC>
|
||||
std::tuple<ck::bhalf_t, NDHWGK, GKZYXC, NDHWGC>>;
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdDataXdl2d : public TestGroupedConvndBwdDataXdl<Tuple>
|
||||
class TestGroupedConvndBwdData2d : public TestGroupedConvndBwdData<Tuple>
|
||||
{
|
||||
};
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdDataXdl3d : public TestGroupedConvndBwdDataXdl<Tuple>
|
||||
class TestGroupedConvndBwdData3d : public TestGroupedConvndBwdData<Tuple>
|
||||
{
|
||||
};
|
||||
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdDataXdl2d, KernelTypes2d);
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdDataXdl3d, KernelTypes3d);
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdData2d, KernelTypes2d);
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdData3d, KernelTypes3d);
|
||||
|
||||
TYPED_TEST(TestGroupedConvndBwdDataXdl2d, Test2D)
|
||||
TYPED_TEST(TestGroupedConvndBwdData2d, Test2D)
|
||||
{
|
||||
this->conv_params.clear();
|
||||
// SplitN case
|
||||
@@ -101,7 +101,7 @@ TYPED_TEST(TestGroupedConvndBwdDataXdl2d, Test2D)
|
||||
this->template Run<2>();
|
||||
}
|
||||
|
||||
TYPED_TEST(TestGroupedConvndBwdDataXdl3d, Test3D)
|
||||
TYPED_TEST(TestGroupedConvndBwdData3d, Test3D)
|
||||
{
|
||||
this->conv_params.clear();
|
||||
// SplitN case
|
||||
@@ -1,135 +0,0 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#include <cstdlib>
|
||||
#include <iostream>
|
||||
#include <initializer_list>
|
||||
#include <tuple>
|
||||
#include <vector>
|
||||
|
||||
#include <gtest/gtest.h>
|
||||
|
||||
#include "profiler/profile_grouped_conv_bwd_data_impl.hpp"
|
||||
|
||||
static ck::index_t param_mask = 0xffff;
|
||||
static ck::index_t instance_index = -1;
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdDataWmma : public ::testing::Test
|
||||
{
|
||||
protected:
|
||||
using DataType = std::tuple_element_t<0, Tuple>;
|
||||
using OutLayout = std::tuple_element_t<1, Tuple>;
|
||||
using WeiLayout = std::tuple_element_t<2, Tuple>;
|
||||
using InLayout = std::tuple_element_t<3, Tuple>;
|
||||
|
||||
std::vector<ck::utils::conv::ConvParam> conv_params;
|
||||
|
||||
template <ck::index_t NDimSpatial>
|
||||
void Run()
|
||||
{
|
||||
EXPECT_FALSE(conv_params.empty());
|
||||
bool pass = true;
|
||||
for(size_t i = 0; i < conv_params.size(); i++)
|
||||
{
|
||||
if((param_mask & (1 << i)) == 0)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
auto& param = conv_params[i];
|
||||
pass = pass && ck::profiler::profile_grouped_conv_bwd_data_impl<NDimSpatial,
|
||||
OutLayout,
|
||||
WeiLayout,
|
||||
InLayout,
|
||||
DataType,
|
||||
DataType,
|
||||
DataType>(
|
||||
true, // do_verification
|
||||
1, // init_method: integer value
|
||||
false, // do_log
|
||||
false, // time_kernel
|
||||
param,
|
||||
1, // splitK
|
||||
instance_index);
|
||||
}
|
||||
EXPECT_TRUE(pass);
|
||||
}
|
||||
};
|
||||
|
||||
using namespace ck::tensor_layout::convolution;
|
||||
|
||||
using KernelTypes2d = ::testing::Types<std::tuple<ck::half_t, GNHWK, GKYXC, GNHWC>,
|
||||
std::tuple<int8_t, GNHWK, GKYXC, GNHWC>,
|
||||
std::tuple<ck::half_t, NHWGK, GKYXC, NHWGC>,
|
||||
std::tuple<int8_t, NHWGK, GKYXC, NHWGC>>;
|
||||
|
||||
using KernelTypes3d = ::testing::Types<std::tuple<ck::half_t, GNDHWK, GKZYXC, GNDHWC>,
|
||||
std::tuple<int8_t, GNDHWK, GKZYXC, GNDHWC>,
|
||||
std::tuple<ck::half_t, NDHWGK, GKZYXC, NDHWGC>,
|
||||
std::tuple<int8_t, NDHWGK, GKZYXC, NDHWGC>>;
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdDataWmma2d : public TestGroupedConvndBwdDataWmma<Tuple>
|
||||
{
|
||||
};
|
||||
|
||||
template <typename Tuple>
|
||||
class TestGroupedConvndBwdDataWmma3d : public TestGroupedConvndBwdDataWmma<Tuple>
|
||||
{
|
||||
};
|
||||
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdDataWmma2d, KernelTypes2d);
|
||||
TYPED_TEST_SUITE(TestGroupedConvndBwdDataWmma3d, KernelTypes3d);
|
||||
|
||||
TYPED_TEST(TestGroupedConvndBwdDataWmma2d, Test2D)
|
||||
{
|
||||
this->conv_params.clear();
|
||||
|
||||
this->conv_params.push_back(
|
||||
{2, 2, 4, 192, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{2, 2, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
|
||||
this->conv_params.push_back(
|
||||
{2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
|
||||
this->conv_params.push_back({2, 1, 1, 1, 32, {8, 8}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
|
||||
this->conv_params.push_back({2, 1, 1, 64, 3, {8, 8}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
|
||||
this->conv_params.push_back({2, 1, 1, 1, 1, {8, 8}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
|
||||
this->template Run<2>();
|
||||
}
|
||||
|
||||
TYPED_TEST(TestGroupedConvndBwdDataWmma3d, Test3D)
|
||||
{
|
||||
this->conv_params.clear();
|
||||
this->conv_params.push_back(
|
||||
{3, 2, 16, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
|
||||
this->conv_params.push_back(
|
||||
{3, 2, 2, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 2, 32, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 1, 32, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 64, 3, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->conv_params.push_back(
|
||||
{3, 1, 1, 1, 1, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
|
||||
this->template Run<3>();
|
||||
}
|
||||
|
||||
int main(int argc, char** argv)
|
||||
{
|
||||
testing::InitGoogleTest(&argc, argv);
|
||||
if(argc == 1) {}
|
||||
else if(argc == 3)
|
||||
{
|
||||
param_mask = strtol(argv[1], nullptr, 0);
|
||||
instance_index = atoi(argv[2]);
|
||||
}
|
||||
else
|
||||
{
|
||||
std::cout << "Usage of " << argv[0] << std::endl;
|
||||
std::cout << "Arg1,2: param_mask instance_index(-1 means all)" << std::endl;
|
||||
}
|
||||
return RUN_ALL_TESTS();
|
||||
}
|
||||
Reference in New Issue
Block a user